Computationally efficient adaptive color correction

To obtain a photo that reproduces the original scene as accurately as possible, it is necessary to solve the problem of color correction, that is, to find a mapping that translates the coordinates of the camera color space (RGB) into the coordinates of the human color space (CIE XYZ). In this article, we consider color correction using lookup tables, pre-built for various lighting conditions. This approach allows you to achieve high speed and accuracy when applying color correction on the device, but requires large amounts of RAM, which, for example, mobile phones do not have. We propose a method for automatic thinning of a set of search tables without loss of accuracy of color correction. The method is based on clustering of the mappings that specify the color correction. To compare the mappings, we propose a criterion for their similarity based on the maximum difference of the generated colors in the target space of a standard CIE XYZ observer. For the proposed criterion, the article provides an effective calculation method and, together with a theorem justifying the correctness of the method.